DeepSeek V4 Pro
DeepSeek-V4-Pro via Fireworks: flagship open-source 1.6T MoE model for frontier reasoning, advanced coding, and long-context agentic workflows. 1M context. Function calling supported. Uses 1 premium request per send before length multipliers. Does not support web search or image input.
You always get the exact model you pick — we never silently route you to another.
About DeepSeek V4 Pro
DeepSeek V4 Pro makes a compelling case that frontier-class coding performance and a one-million-token context window do not have to cost frontier-class money. At roughly $0.18 per million tokens blended, it runs 10x cheaper on input and 30x cheaper on output than comparable models, while posting an 80.6% score on SWE-Bench Verified — the highest reported among open-weight models at launch. Users consistently praise its agentic coding ability, noting it competes with or beats larger closed models on multi-step coding tasks, and its hybrid attention architecture handles full-codebase analysis without collapsing under the token budget. The MIT license is a genuine differentiator: weights are freely available for self-hosting, fine-tuning, and commercial integration. The honest caveat: V4 Pro is verbose. It can generate four to five times more output tokens than comparable models on the same prompt, which erodes the per-token savings and makes cost estimation harder than it first appears. Still in preview as of mid-2026, with all benchmark scores currently vendor-reported, it is best suited for teams comfortable with that tradeoff.
Best for
- High-volume automated coding pipelines, code review, and refactoring where per-call cost matters
- Full-codebase analysis using 1M-token context for migration planning or architectural review
- Multi-turn agentic workflows where reasoning must persist across tool calls and conversation turns
- Long-document synthesis, research corpora summarization, and extraction from large structured data
- On-premises or self-hosted deployment via MIT-licensed weights with custom fine-tuning
Specifications
| Provider | DeepSeek |
|---|---|
| Released | 2026-04-24 |
| Intelligence | High |
| Speed | Medium |
| Context window | 1,048,600 tokens |
| Max output | 384,000 tokens |
| Knowledge cutoff | April 2026 |
| Input price | $1.74 / 1M tokens |
| Output price | $3.48 / 1M tokens |
| Request cost | 1 premium request |
| Plan tier | Premium |
| Input | Text |
| Output | Text |
| Features | Cached input: $0.14 / 1M tokens, 1 premium request per send before length multipliers, Function calling supported, Fine-tuning not supported on Fireworks serverless, On-demand Fireworks deployments available separately |
| Model ID | deepseek-v4-pro |
Frequently asked questions
Input is $0.435 per million tokens and output is $0.87 per million tokens, with a 99% discount on cache hits ($0.003625/M). The blended effective price is roughly $0.18/M tokens. DeepSeek made a 75% discount permanent in May 2026.
1,048,576 tokens (approximately 1 million tokens), with a maximum output of 384,000 tokens.
Not yet as of mid-2026. All published scores — including 80.6% SWE-Bench Verified and 90.1% GPQA Diamond — come from DeepSeek's internal evaluations. Independent third-party leaderboard verification is pending.
It produces significantly more output tokens than most models on the same prompts — sometimes 4 to 5 times more — making real costs harder to predict. It also has a reliability gap versus top closed models on the most complex edge-case reasoning tasks, and it censors politically sensitive topics related to Chinese governance.
Teams running cost-sensitive, high-volume coding or document workloads who can self-host or accept the API's privacy trade-offs, and who want MIT-licensed weights for fine-tuning or commercial integration.
It remains a preview release as of June 2026. DeepSeek has not announced a stable release date.